[an error occurred while processing this directive] [an error occurred while processing this directive] [an error occurred while processing this directive] [an error occurred while processing this directive] [an error occurred while processing this directive] [an error occurred while processing this directive] [an error occurred while processing this directive] [an error occurred while processing this directive] [an error occurred while processing this directive] [an error occurred while processing this directive] [an error occurred while processing this directive] /Opinnot/T-61.6090/2007/projectwork.shtml [an error occurred while processing this directive] [an error occurred while processing this directive] [an error occurred while processing this directive]
Laboratory of Computer and Information Science / Neural Networks Research Centre CIS Lab Helsinki University of Technology

Project Work

General Instructions

The group signs up by emailing the names of group members, the chosen topic, and their preferred project presentation day (5.11. or 26.11.) to the course address by Mon 15 Oct. Once the groups have signed up, they will be announced on the course web page.

The deadline for project reports and literature essays is Mon 3 Dec. The returned reports and essays will then be made available for course participants on the internal webpages.

Schedule

Topics

1. Group work: Single confabulation for language generation

This project is preferably performed as a group excercise (2-3 people).

Build a word predictor like the one described in Hecht-Nielsen's book in chapter 7.3.1. Train it using a suitable large english training corpus. Generate sentence continuations for some sentences. Corpora availability can be complicated due to licensing issues. Since many course participants have access to large corpora of english text that are not publicly available, we suggest that groups are formed so that someone has access to a large corpus. If some group does not have access to any large corpora, a backup plan can be using freely available corpora, such as Grimm's Fairy Tales from project Gutenberg, but note that it is rather small.

Build a simple n-gram model as a baseline comparison. Try generating completions from it in similar ways. Which one do your think performs better? How could it be assessed in a more rigorous fashion? On what kinds of data do the models work better? For example: How does sentence length affect the different completions? We recommend using a third party software package to build and train the n-gram model. For example SRILM-toolkit (http://www.speech.sri.com/projects/srilm/). You could also use the recently developed VariKN language modeling toolkit by Vesa Siivola (http://www.cis.hut.fi/vsiivola/).

Return a report on the experiments and results that is 5-10 pages long where you discuss your expectations, the experiment, and your results. How could the model be improved, either within the framework of confabulation theory, or outside?

Also describe your experience of building the model. What did you learn? You should also briefly describe your opinions about confabulation theory as a theory of cognition, and if doing this project affected your opinion.

Each group gives a 30 min presentation at a seminar day. Every member of the group should participate in giving the group presentation.

2. Individual work: Literature study

This is suitable for working individually or in pairs. If you would like to do it pairwise, by all means discuss the topic together, but each of you should finally return your own essay.

Each person gives a presentation of 15-20 minutes during one of the seminar days.

Choose a topic that genuinely interests you, and select source materials from course readings from at least two authors that were related to that topic. In many cases there's no need to cover all the material by each author. If you consider it relevant, you can additionally use material outside of course readings, even to include additional authors.

Then, write an essay of 8-15 pages:

You are also welcome to include discussion of what you learned in doing this, other comments or suggestions regarding this project work, or on the course in general.

3. Your own topic

Suggest your own topic. It should be a practical project like the confabulation experiment.

Example description:

Modeling the state of the world using Temporal Bayesian Networks

Model the state of the world in the princess story below as accurately as possible using temporal bayesian networks. It's ok to only model a few paragraphs of the story, the point is more to consider the problem, than build an exhaustive model.

"In olden times when wishing still helped one, there lived a king whose daughters were all beautiful, but the youngest was so beautiful that the sun itself, which has seen so much, was astonished whenever it shone in her face. Close by the king's castle lay a great dark forest, and under an old lime-tree in the forest was a well, and when the day was very warm, the king's child went out into the forest and sat down by the side of the cool fountain, and when she was bored she took a golden ball, and threw it up on high and caught it, and this ball was her favorite plaything. ..." from the Grimm fairy tales

Discuss how well you can model the world in this way. Is it realistic from a cognitive viewpoint? How well can you reason about things not explicitly mentioned in the text.

You are at: CIS → /Opinnot/T-61.6090/2007/projectwork.shtml

Page maintained by webmaster at cis.hut.fi, last updated Monday, 08-Oct-2007 16:14:27 EEST

[an error occurred while processing this directive]
Google
WWW www.cis.hut.fi